Features
Built with serious AI for serious education
Multi-agent architecture, vision processing, RAG retrieval, and pedagogical strategies — all working together to create the best AI tutoring experience.
Multi-Agent Architecture
Five specialized agents handle different question types: syllabus policies, lecture content, assignment tutoring, professor analytics, and intelligent routing. Each agent has deep domain expertise.
- Task Manager routes queries to the right specialist
- Syllabus Agent handles policies, deadlines, and logistics
- Content Agent explains concepts with step-by-step teaching
- Assignments Agent tutors with configurable strategies
- Professor Agent provides analytics and insights
Visual / diagram placeholder
RAG-Powered Accuracy
Every answer is grounded in your actual course materials using Retrieval-Augmented Generation. No hallucination, no generic internet answers — just your content.
- FAISS vector indexing for millisecond retrieval
- OpenAI embeddings for semantic search
- Paragraph-based chunking with overlap
- Source citations in every response
- Supports PDF, DOCX, PPTX, HTML, and more
Visual / diagram placeholder
Vision Processing
Our vision pipeline extracts and understands images, formulas, diagrams, and handwritten work using GPT-4o or Gemini.
- Reads handwritten solutions and formulas
- Extracts images and diagrams from PDFs
- Math font detection (CMMI, CMSY, CMR)
- High-DPI rendering for accurate transcription
- Students can search for visual content
Visual / diagram placeholder
Pedagogical Tutoring
Three research-backed teaching strategies ensure students learn through guided discovery, never by being handed answers.
- Socratic Navigator: Question-first approach
- Hints Ladder: Progressive from general to specific
- Metacognitive Coach: Reflection and self-awareness
- Never gives direct answers (hard rule)
- Encourages self-explanation and reasoning
Visual / diagram placeholder
Professor Analytics
A dedicated analytics agent analyzes all student interactions to give professors actionable insights about their course.
- Engagement metrics and participation rates
- Confusion hotspot identification
- Question categorization and patterns
- Individual student learning profiles
- Data-driven improvement recommendations
Visual / diagram placeholder
Privacy & Security
FERPA and GDPR compliant by design. Student identifiers are hashed, no PII is stored, and professor features are role-gated.
- SHA-256 hashed student identifiers
- No personally identifiable information stored
- Role-based access control
- Students cannot access professor features
- Secure session management via Supabase
Visual / diagram placeholder
Under the Hood
Technical specifications
AI Models
GPT-4o, GPT-4o-mini, Gemini
Vector Search
FAISS with OpenAI embeddings
Document Types
PDF, DOCX, PPTX, HTML, images
Solution Upload
PDF, DOCX, JPEG, PNG
Query Speed
2-5 seconds
Scalability
100+ PDFs per course
Compliance
FERPA & GDPR
Agents
5 specialized + routing
FAQ
Frequently asked questions
Everything you need to know about Thalia.
Never. Thalia is designed with strict pedagogical rules — it uses Socratic questioning, progressive hints, and metacognitive coaching to guide students toward understanding. Giving direct answers is explicitly forbidden in the system.
Thalia works with any course that has digital materials. It's especially powerful for STEM courses with formulas, diagrams, and problem sets. Current deployments include electrical engineering, computer science, and machine learning courses.
Thalia ingests your course materials (PDFs, slides, documents) and creates a searchable knowledge base using FAISS vector indexing. When a student asks a question, the system retrieves relevant snippets from your actual materials to ground every response.
Absolutely. Thalia is FERPA and GDPR compliant. Student identifiers are SHA-256 hashed before storage. No personally identifiable information is ever stored in our analytics database.
Yes. Students can upload photos or scans of handwritten solutions, and our vision-powered system (using GPT-4o) can read and analyze their work — including formulas, diagrams, and circuit drawings.
Most courses can be indexed in under 15 minutes. Upload your materials, run the ingestion pipeline, and your AI tutor is ready. Vision processing for images and formulas adds a small one-time cost.
Ready to transform your classroom?
Join educators who are using AI to create better learning experiences — without sacrificing academic integrity.